Robotic Process Automation (RPA) and similar digital technologies are currently being implemented for automation of work processes in public service provision. Although RPA enables new empirical phenomena in the digital government context, automation of work is not a new and understudied phenomenon. In this paper, I claim that there is a risk that digital government researchers treat this phenomenon as being entirely new, omitting years of experiences made in other, related, fields of research. To prevent this risk, I call for digital government researchers to learn from adjacent research disciplines. I follow this call and present four ironies of automation, extracted from Lisanne Bainbridges iconic work from 1983, and relate these to the public service automation context, using examples of RPA in local government organizations. The ironies concern contradictions in the underlying views on humans and human error in the design of automated systems and human-automation interaction, as well as the consequences of human-automation interaction in terms of new tasks (monitoring and take-over), re-configurations of responsibilities, and de- and re-skilling of humans interacting with automated systems. These findings can guide digital government theorization and empirical research on public service automation. Learning from adjacent fields of research is important to better understand and cumulatively build knowledge on the characteristics and effects of public service automation. Learning from an established knowledge base is also important to avoid foreseeable project and implementation failures. The paper is concluded with suggestions on future research topics for digital government researchers interested in public service automation.